人脸生成（Face Generation）¶

在该项目中，你将使用生成式对抗网络（Generative Adversarial Nets）来生成新的人脸图像。

获取数据¶

该项目将使用以下数据集：

MNIST
CelebA

由于 CelebA 数据集比较复杂，而且这是你第一次使用 GANs。我们想让你先在 MNIST 数据集上测试你的 GANs 模型，以让你更快的评估所建立模型的性能。

如果你在使用 FloydHub, 请将 data_dir 设置为 "/input" 并使用 FloydHub data ID "R5KrjnANiKVhLWAkpXhNBe".

data_dir = './data'

# FloydHub - Use with data ID "R5KrjnANiKVhLWAkpXhNBe"
#data_dir = '/input'


"""
DON'T MODIFY ANYTHING IN THIS CELL
"""
import helper

helper.download_extract('mnist', data_dir)
helper.download_extract('celeba', data_dir)
#helper.download_extract('celeba', cele_data_dir)

Found mnist Data
Found celeba Data

探索数据（Explore the Data）¶

MNIST¶

MNIST 是一个手写数字的图像数据集。你可以更改 show_n_images 探索此数据集。

show_n_images = 25

"""
DON'T MODIFY ANYTHING IN THIS CELL
"""
%matplotlib inline
import os
from glob import glob
from matplotlib import pyplot

mnist_images = helper.get_batch(glob(os.path.join(data_dir, 'mnist/*.jpg'))[:show_n_images], 28, 28, 'L')
pyplot.imshow(helper.images_square_grid(mnist_images, 'L'), cmap='gray')

<matplotlib.image.AxesImage at 0x1a8a730ef98>

CelebA¶

CelebFaces Attributes Dataset (CelebA) 是一个包含 20 多万张名人图片及相关图片说明的数据集。你将用此数据集生成人脸，不会用不到相关说明。你可以更改 show_n_images 探索此数据集。

show_n_images = 25

"""
DON'T MODIFY ANYTHING IN THIS CELL
"""
mnist_images = helper.get_batch(glob(os.path.join(data_dir, 'img_align_celeba/*.jpg'))[:show_n_images], 28, 28, 'RGB')
pyplot.imshow(helper.images_square_grid(mnist_images, 'RGB'))

<matplotlib.image.AxesImage at 0x1a8a74366a0>

预处理数据（Preprocess the Data）¶

由于该项目的重点是建立 GANs 模型，我们将为你预处理数据。

经过数据预处理，MNIST 和 CelebA 数据集的值在 28×28 维度图像的 [-0.5, 0.5] 范围内。CelebA 数据集中的图像裁剪了非脸部的图像部分，然后调整到 28x28 维度。

MNIST 数据集中的图像是单通道的黑白图像，CelebA 数据集中的图像是三通道的 RGB 彩色图像。

建立神经网络（Build the Neural Network）¶

你将通过部署以下函数来建立 GANs 的主要组成部分:

model_inputs
discriminator
generator
model_loss
model_opt
train

检查 TensorFlow 版本并获取 GPU 型号¶

检查你是否使用正确的 TensorFlow 版本，并获取 GPU 型号

"""
DON'T MODIFY ANYTHING IN THIS CELL
"""
from distutils.version import LooseVersion
import warnings
import tensorflow as tf

# Check TensorFlow Version
assert LooseVersion(tf.__version__) >= LooseVersion('1.0'), 'Please use TensorFlow version 1.0 or newer.  You are using {}'.format(tf.__version__)
print('TensorFlow Version: {}'.format(tf.__version__))

# Check for a GPU
if not tf.test.gpu_device_name():
    warnings.warn('No GPU found. Please use a GPU to train your neural network.')
else:
    print('Default GPU Device: {}'.format(tf.test.gpu_device_name()))

TensorFlow Version: 1.2.0-rc1
Default GPU Device: /gpu:0

输入（Input）¶

部署 model_inputs 函数以创建用于神经网络的占位符 (TF Placeholders)。请创建以下占位符：

输入图像占位符: 使用 image_width，image_height 和 image_channels 设置为 rank 4。
输入 Z 占位符: 设置为 rank 2，并命名为 z_dim。
学习速率占位符: 设置为 rank 0。

返回占位符元组的形状为 (tensor of real input images, tensor of z data, learning rate)。

import problem_unittests as tests

def model_inputs(image_width, image_height, image_channels, z_dim):
    """
    Create the model inputs
    :param image_width: The input image width
    :param image_height: The input image height
    :param image_channels: The number of image channels
    :param z_dim: The dimension of Z
    :return: Tuple of (tensor of real input images, tensor of z data, learning rate)
    """
    # TODO: Implement Function
    image = tf.placeholder(tf.float32, [None, image_width, image_height, image_channels], name = 'real_input')
    z = tf.placeholder(tf.float32, [None, z_dim], name = 'z_dim')
    lr = tf.placeholder(tf.float32, name = 'learning_rate')
    return image, z, lr


"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
tests.test_model_inputs(model_inputs)

ERROR:tensorflow:==================================
Object was never used (type <class 'tensorflow.python.framework.ops.Operation'>):
<tf.Operation 'assert_rank_2/Assert/Assert' type=Assert>
If you want to mark it as used call its "mark_used()" method.
It was originally created here:
['File "D:\\Anaconda3\\lib\\runpy.py", line 193, in _run_module_as_main\n    "__main__", mod_spec)', 'File "D:\\Anaconda3\\lib\\runpy.py", line 85, in _run_code\n    exec(code, run_globals)', 'File "D:\\Anaconda3\\lib\\site-packages\\ipykernel_launcher.py", line 16, in <module>\n    app.launch_new_instance()', 'File "D:\\Anaconda3\\lib\\site-packages\\traitlets\\config\\application.py", line 658, in launch_instance\n    app.start()', 'File "D:\\Anaconda3\\lib\\site-packages\\ipykernel\\kernelapp.py", line 477, in start\n    ioloop.IOLoop.instance().start()', 'File "D:\\Anaconda3\\lib\\site-packages\\zmq\\eventloop\\ioloop.py", line 177, in start\n    super(ZMQIOLoop, self).start()', 'File "D:\\Anaconda3\\lib\\site-packages\\tornado\\ioloop.py", line 888, in start\n    handler_func(fd_obj, events)', 'File "D:\\Anaconda3\\lib\\site-packages\\tornado\\stack_context.py", line 277, in null_wrapper\n    return fn(*args, **kwargs)', 'File "D:\\Anaconda3\\lib\\site-packages\\zmq\\eventloop\\zmqstream.py", line 440, in _handle_events\n    self._handle_recv()', 'File "D:\\Anaconda3\\lib\\site-packages\\zmq\\eventloop\\zmqstream.py", line 472, in _handle_recv\n    self._run_callback(callback, msg)', 'File "D:\\Anaconda3\\lib\\site-packages\\zmq\\eventloop\\zmqstream.py", line 414, in _run_callback\n    callback(*args, **kwargs)', 'File "D:\\Anaconda3\\lib\\site-packages\\tornado\\stack_context.py", line 277, in null_wrapper\n    return fn(*args, **kwargs)', 'File "D:\\Anaconda3\\lib\\site-packages\\ipykernel\\kernelbase.py", line 283, in dispatcher\n    return self.dispatch_shell(stream, msg)', 'File "D:\\Anaconda3\\lib\\site-packages\\ipykernel\\kernelbase.py", line 235, in dispatch_shell\n    handler(stream, idents, msg)', 'File "D:\\Anaconda3\\lib\\site-packages\\ipykernel\\kernelbase.py", line 399, in execute_request\n    user_expressions, allow_stdin)', 'File "D:\\Anaconda3\\lib\\site-packages\\ipykernel\\ipkernel.py", line 196, in do_execute\n    res = shell.run_cell(code, store_history=store_history, silent=silent)', 'File "D:\\Anaconda3\\lib\\site-packages\\ipykernel\\zmqshell.py", line 533, in run_cell\n    return super(ZMQInteractiveShell, self).run_cell(*args, **kwargs)', 'File "D:\\Anaconda3\\lib\\site-packages\\IPython\\core\\interactiveshell.py", line 2698, in run_cell\n    interactivity=interactivity, compiler=compiler, result=result)', 'File "D:\\Anaconda3\\lib\\site-packages\\IPython\\core\\interactiveshell.py", line 2808, in run_ast_nodes\n    if self.run_code(code, result):', 'File "D:\\Anaconda3\\lib\\site-packages\\IPython\\core\\interactiveshell.py", line 2862, in run_code\n    exec(code_obj, self.user_global_ns, self.user_ns)', 'File "<ipython-input-5-cc35e4ed6ed5>", line 22, in <module>\n    tests.test_model_inputs(model_inputs)', 'File "D:\\学习\\Udacity_深度学习\\projects\\face-generation\\problem_unittests.py", line 12, in func_wrapper\n    result = func(*args)', 'File "D:\\学习\\Udacity_深度学习\\projects\\face-generation\\problem_unittests.py", line 68, in test_model_inputs\n    _check_input(learn_rate, [], \'Learning Rate\')', 'File "D:\\学习\\Udacity_深度学习\\projects\\face-generation\\problem_unittests.py", line 34, in _check_input\n    _assert_tensor_shape(tensor, shape, \'Real Input\')', 'File "D:\\学习\\Udacity_深度学习\\projects\\face-generation\\problem_unittests.py", line 20, in _assert_tensor_shape\n    assert tf.assert_rank(tensor, len(shape), message=\'{} has wrong rank\'.format(display_name))', 'File "D:\\Anaconda3\\lib\\site-packages\\tensorflow\\python\\ops\\check_ops.py", line 617, in assert_rank\n    dynamic_condition, data, summarize)', 'File "D:\\Anaconda3\\lib\\site-packages\\tensorflow\\python\\ops\\check_ops.py", line 571, in _assert_rank_condition\n    return control_flow_ops.Assert(condition, data, summarize=summarize)', 'File "D:\\Anaconda3\\lib\\site-packages\\tensorflow\\python\\util\\tf_should_use.py", line 170, in wrapped\n    return _add_should_use_warning(fn(*args, **kwargs))', 'File "D:\\Anaconda3\\lib\\site-packages\\tensorflow\\python\\util\\tf_should_use.py", line 139, in _add_should_use_warning\n    wrapped = TFShouldUseWarningWrapper(x)', 'File "D:\\Anaconda3\\lib\\site-packages\\tensorflow\\python\\util\\tf_should_use.py", line 96, in __init__\n    stack = [s.strip() for s in traceback.format_stack()]']
==================================
Tests Passed

辨别器（Discriminator）¶

部署 discriminator 函数创建辨别器神经网络以辨别 images。该函数应能够重复使用神经网络中的各种变量。在 tf.variable_scope 中使用 "discriminator" 的变量空间名来重复使用该函数中的变量。

该函数应返回形如 (tensor output of the discriminator, tensor logits of the discriminator) 的元组。

def discriminator(images, reuse=False, alpha = 0.2):
    """
    Create the discriminator network
    :param image: Tensor of input image(s)
    :param reuse: Boolean if the weights should be reused
    :return: Tuple of (tensor output of the discriminator, tensor logits of the discriminator)
    """
    # TODO: Implement Function
    with tf.variable_scope('discriminator', reuse=reuse):
        # Input layer is 28x28x3 or 28x28x1
        x1 = tf.layers.conv2d(images, 128, 5, strides=2, padding='SAME', kernel_initializer= tf.contrib.layers.xavier_initializer())
        relu1 = tf.maximum(alpha * x1, x1)
        relu1 = tf.nn.dropout(relu1, 0.8)
        #print(x1.get_shape())
        # 14x14x128
        
        x2 = tf.layers.conv2d(relu1, 256, 5, strides=2, padding='SAME', kernel_initializer= tf.contrib.layers.xavier_initializer())
        bn2 = tf.layers.batch_normalization(x2, training=True)
        relu2 = tf.maximum(alpha * bn2, bn2)
        relu2 = tf.nn.dropout(relu2, 0.8)
        #print(x2.get_shape())
        # 7x7x256
        
        x3 = tf.layers.conv2d(relu2, 512, 5, strides=2, padding='SAME', kernel_initializer= tf.contrib.layers.xavier_initializer())
        bn3 = tf.layers.batch_normalization(x3, training=True)
        relu3 = tf.maximum(alpha * bn3, bn3)
        relu3 = tf.nn.dropout(relu3, 0.8)
        # 4x4x512
        #print(x3.get_shape())

        # Flatten it
        flat = tf.reshape(relu3, (-1, 4*4*512))
        logits = tf.layers.dense(flat, 1)
        output = tf.sigmoid(logits)
        
        
        return output, logits


"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
tests.test_discriminator(discriminator, tf)

Tests Passed

生成器（Generator）¶

部署 generator 函数以使用 z 生成图像。该函数应能够重复使用神经网络中的各种变量。在 tf.variable_scope 中使用 "generator" 的变量空间名来重复使用该函数中的变量。

该函数应返回所生成的 28 x 28 x out_channel_dim 维度图像。

def generator(z, out_channel_dim, is_train=True, alpha = 0.2):
    """
    Create the generator network
    :param z: Input z
    :param out_channel_dim: The number of channels in the output image
    :param is_train: Boolean if generator is being used for training
    :return: The tensor output of the generator
    """
    # TODO: Implement Function
    with tf.variable_scope('generator',reuse = not is_train):
        x1 = tf.layers.dense(z, 7*7*512)
        
        # Reshape it to start the convolutional stack
        x1 = tf.reshape(x1, (-1, 7, 7, 512))
        x1 = tf.layers.batch_normalization(x1, training=is_train)
        x1 = tf.maximum(alpha * x1, x1)
        x1 = tf.nn.dropout(x1, 0.8)
        #print(x1.get_shape())
        # 7x7x512 now
        
        x2 = tf.layers.conv2d_transpose(x1, 256, 5, strides=2, padding='SAME', kernel_initializer = tf.contrib.layers.xavier_initializer())
        x2 = tf.layers.batch_normalization(x2, training=is_train)
        x2 = tf.maximum(alpha * x2, x2)
        #print(x2.get_shape())
        # 14x14x256 now
        
        x3 = tf.layers.conv2d_transpose(x2, 128, 5, strides=2, padding='SAME', kernel_initializer = tf.contrib.layers.xavier_initializer())
        x3 = tf.layers.batch_normalization(x3, training=is_train)
        x3 = tf.maximum(alpha * x3, x3)
        #print(x3.get_shape())
        # 28x28x128 now
        
        # Output layer
        logits = tf.layers.conv2d_transpose(x3, out_channel_dim, 5, strides=1, padding='SAME', 
                                            kernel_initializer = tf.contrib.layers.xavier_initializer())
        # 28x28x3 now
        
        output = tf.tanh(logits)     
        
        
        return output * 0.5


"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
tests.test_generator(generator, tf)

Tests Passed

损失函数（Loss）¶

部署 model_loss 函数训练并计算 GANs 的损失。该函数应返回形如 (discriminator loss, generator loss) 的元组。

使用你已实现的函数：

discriminator(images, reuse=False)
generator(z, out_channel_dim, is_train=True)

def model_loss(input_real, input_z, out_channel_dim):
    """
    Get the loss for the discriminator and generator
    :param input_real: Images from the real dataset
    :param input_z: Z input
    :param out_channel_dim: The number of channels in the output image
    :return: A tuple of (discriminator loss, generator loss)
    """
    # TODO: Implement Function
    g_model = generator(input_z, out_channel_dim)
    d_model_real, d_logits_real = discriminator(input_real)
    d_model_fake, d_logits_fake = discriminator(g_model, reuse=True)

    d_loss_real = tf.reduce_mean(
        tf.nn.sigmoid_cross_entropy_with_logits(logits=d_logits_real, labels=tf.ones_like(d_model_real) * 0.9))
    d_loss_fake = tf.reduce_mean(
        tf.nn.sigmoid_cross_entropy_with_logits(logits=d_logits_fake, labels=tf.zeros_like(d_model_fake)))
    g_loss = tf.reduce_mean(
        tf.nn.sigmoid_cross_entropy_with_logits(logits=d_logits_fake, labels=tf.ones_like(d_model_fake)))

    d_loss = d_loss_real + d_loss_fake
    
    return g_loss, d_loss


"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
tests.test_model_loss(model_loss)

Tests Passed

优化（Optimization）¶

部署 model_opt 函数实现对 GANs 的优化。使用 tf.trainable_variables 获取可训练的所有变量。通过变量空间名 discriminator 和 generator 来过滤变量。该函数应返回形如 (discriminator training operation, generator training operation) 的元组。

def model_opt(d_loss, g_loss, learning_rate, beta1):
    """
    Get optimization operations
    :param d_loss: Discriminator loss Tensor
    :param g_loss: Generator loss Tensor
    :param learning_rate: Learning Rate Placeholder
    :param beta1: The exponential decay rate for the 1st moment in the optimizer
    :return: A tuple of (discriminator training operation, generator training operation)
    """
    # TODO: Implement Function
    t_vars = tf.trainable_variables()
    d_vars = [var for var in t_vars if var.name.startswith('discriminator')]
    g_vars = [var for var in t_vars if var.name.startswith('generator')]

    # Optimize
    with tf.control_dependencies(tf.get_collection(tf.GraphKeys.UPDATE_OPS)):
        d_train_opt = tf.train.AdamOptimizer(learning_rate, beta1=beta1).minimize(d_loss, var_list=d_vars)
        g_train_opt = tf.train.AdamOptimizer(learning_rate, beta1=beta1).minimize(g_loss, var_list=g_vars)
        
    return d_train_opt, g_train_opt


"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
tests.test_model_opt(model_opt, tf)

Tests Passed

训练神经网络（Neural Network Training）¶

输出显示¶

使用该函数可以显示生成器 (Generator) 在训练过程中的当前输出，这会帮你评估 GANs 模型的训练程度。

"""
DON'T MODIFY ANYTHING IN THIS CELL
"""
import numpy as np

def show_generator_output(sess, n_images, input_z, out_channel_dim, image_mode):
    """
    Show example output for the generator
    :param sess: TensorFlow session
    :param n_images: Number of Images to display
    :param input_z: Input Z Tensor
    :param out_channel_dim: The number of channels in the output image
    :param image_mode: The mode to use for images ("RGB" or "L")
    """
    cmap = None if image_mode == 'RGB' else 'gray'
    z_dim = input_z.get_shape().as_list()[-1]
    example_z = np.random.uniform(-1, 1, size=[n_images, z_dim])

    samples = sess.run(
        generator(input_z, out_channel_dim, False),
        feed_dict={input_z: example_z})

    images_grid = helper.images_square_grid(samples, image_mode)
    pyplot.imshow(images_grid, cmap=cmap)
    pyplot.show()

训练¶

部署 train 函数以建立并训练 GANs 模型。记得使用以下你已完成的函数：

model_inputs(image_width, image_height, image_channels, z_dim)
model_loss(input_real, input_z, out_channel_dim)
model_opt(d_loss, g_loss, learning_rate, beta1)

使用 show_generator_output 函数显示 generator 在训练过程中的输出。

注意：在每个批次 (batch) 中运行 show_generator_output 函数会显著增加训练时间与该 notebook 的体积。推荐每 100 批次输出一次 generator 的输出。

def train(epoch_count, batch_size, z_dim, learning_rate, beta1, get_batches, data_shape, data_image_mode):
    """
    Train the GAN
    :param epoch_count: Number of epochs
    :param batch_size: Batch Size
    :param z_dim: Z dimension
    :param learning_rate: Learning Rate
    :param beta1: The exponential decay rate for the 1st moment in the optimizer
    :param get_batches: Function to get batches
    :param data_shape: Shape of the data
    :param data_image_mode: The image mode to use for images ("RGB" or "L")
    """
    # TODO: Build Model
    image_width = data_shape[1]
    image_height = data_shape[2]
    image_channels = data_shape[3]
    
    input_real, input_z, lr = model_inputs(image_width, image_height, image_channels, z_dim)
    g_loss, d_loss = model_loss(input_real, input_z, image_channels)
    d_train_opt, g_train_opt = model_opt(d_loss, g_loss, lr, beta1)
    
    loss_d, loss_g = [], []
    
    with tf.Session() as sess:
        sess.run(tf.global_variables_initializer())
        for epoch_i in range(epoch_count):
            step = 0
            for batch_images in get_batches(batch_size):
                # TODO: Train Model
                step += 1
                batch_z = np.random.uniform(-1, 1, size=(batch_size, z_dim)).astype(np.float32)
                # 不知道是什么原因，放大到-1到1后，g_loss变得很大，于是将tanh的输出缩放到-0.5到0.5
                #batch_images = batch_images * 2
                
                # 这里run两次g_train_opt，为了避免辨别器收敛太快，导致生成的图片完全没有能力PK
                # 参考 https://github.com/carpedm20/DCGAN-tensorflow
                
                _ = sess.run(d_train_opt, feed_dict={input_real: batch_images, input_z: batch_z, lr: learning_rate})
                _ = sess.run(g_train_opt, feed_dict={input_real: batch_images, input_z: batch_z, lr: learning_rate})
                _ = sess.run(g_train_opt, feed_dict={input_real: batch_images, input_z: batch_z, lr: learning_rate})
                
                if step % 10 == 0:
                    train_loss_d = d_loss.eval({input_z: batch_z, input_real: batch_images})
                    train_loss_g = g_loss.eval({input_z: batch_z})
                    loss_d.append(train_loss_d)
                    loss_g.append(train_loss_g)
                    print("Epoch {}/{}...".format(epoch_i+1, epoch_count), "Batch {} ".format(step),
                          "Discriminator Loss: {:.4f}...".format(train_loss_d),
                          "Generator Loss: {:.4f}".format(train_loss_g))
                if step % 100 == 0:
                    show_generator_output(sess, 25, input_z, image_channels, data_image_mode)
        return loss_d, loss_g

MNIST¶

在 MNIST 上测试你的 GANs 模型。经过 2 次迭代，GANs 应该能够生成类似手写数字的图像。确保生成器 (generator) 低于辨别器 (discriminator) 的损失，或接近 0。

batch_size = 64
z_dim = 100
learning_rate = 0.0002
beta1 = 0.5


"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
epochs = 2

mnist_dataset = helper.Dataset('mnist', glob(os.path.join(data_dir, 'mnist/*.jpg')))
with tf.Graph().as_default():
    train(epochs, batch_size, z_dim, learning_rate, beta1, mnist_dataset.get_batches,
          mnist_dataset.shape, mnist_dataset.image_mode)

Epoch 1/2... Batch 10  Discriminator Loss: 3.1505... Generator Loss: 0.1574
Epoch 1/2... Batch 20  Discriminator Loss: 0.4022... Generator Loss: 6.2334
Epoch 1/2... Batch 30  Discriminator Loss: 0.3854... Generator Loss: 5.6580
Epoch 1/2... Batch 40  Discriminator Loss: 2.1076... Generator Loss: 0.4358
Epoch 1/2... Batch 50  Discriminator Loss: 1.5201... Generator Loss: 1.8188
Epoch 1/2... Batch 60  Discriminator Loss: 2.7243... Generator Loss: 0.4288
Epoch 1/2... Batch 70  Discriminator Loss: 1.9254... Generator Loss: 0.6638
Epoch 1/2... Batch 80  Discriminator Loss: 0.9059... Generator Loss: 1.2986
Epoch 1/2... Batch 90  Discriminator Loss: 1.8479... Generator Loss: 0.5618
Epoch 1/2... Batch 100  Discriminator Loss: 1.3742... Generator Loss: 0.8914

Epoch 1/2... Batch 110  Discriminator Loss: 1.4778... Generator Loss: 0.7693
Epoch 1/2... Batch 120  Discriminator Loss: 1.4035... Generator Loss: 1.2067
Epoch 1/2... Batch 130  Discriminator Loss: 1.4173... Generator Loss: 1.2127
Epoch 1/2... Batch 140  Discriminator Loss: 1.7627... Generator Loss: 0.4710
Epoch 1/2... Batch 150  Discriminator Loss: 1.8127... Generator Loss: 0.4515
Epoch 1/2... Batch 160  Discriminator Loss: 1.7373... Generator Loss: 0.7082
Epoch 1/2... Batch 170  Discriminator Loss: 1.6547... Generator Loss: 0.6542
Epoch 1/2... Batch 180  Discriminator Loss: 1.7679... Generator Loss: 0.6173
Epoch 1/2... Batch 190  Discriminator Loss: 1.5722... Generator Loss: 0.7403
Epoch 1/2... Batch 200  Discriminator Loss: 1.4693... Generator Loss: 0.6776

Epoch 1/2... Batch 210  Discriminator Loss: 1.2390... Generator Loss: 1.0140
Epoch 1/2... Batch 220  Discriminator Loss: 1.5497... Generator Loss: 0.7767
Epoch 1/2... Batch 230  Discriminator Loss: 1.4413... Generator Loss: 0.6109
Epoch 1/2... Batch 240  Discriminator Loss: 1.5819... Generator Loss: 0.7183
Epoch 1/2... Batch 250  Discriminator Loss: 1.4917... Generator Loss: 0.8840
Epoch 1/2... Batch 260  Discriminator Loss: 1.3610... Generator Loss: 0.8013
Epoch 1/2... Batch 270  Discriminator Loss: 1.3385... Generator Loss: 0.9811
Epoch 1/2... Batch 280  Discriminator Loss: 1.5133... Generator Loss: 0.6366
Epoch 1/2... Batch 290  Discriminator Loss: 1.5110... Generator Loss: 0.6250
Epoch 1/2... Batch 300  Discriminator Loss: 1.3658... Generator Loss: 0.9430

Epoch 1/2... Batch 310  Discriminator Loss: 1.3748... Generator Loss: 0.7377
Epoch 1/2... Batch 320  Discriminator Loss: 1.4442... Generator Loss: 0.6260
Epoch 1/2... Batch 330  Discriminator Loss: 1.3476... Generator Loss: 0.9196
Epoch 1/2... Batch 340  Discriminator Loss: 1.4841... Generator Loss: 0.6621
Epoch 1/2... Batch 350  Discriminator Loss: 1.4058... Generator Loss: 1.0552
Epoch 1/2... Batch 360  Discriminator Loss: 1.5038... Generator Loss: 0.9391
Epoch 1/2... Batch 370  Discriminator Loss: 1.4998... Generator Loss: 0.8521
Epoch 1/2... Batch 380  Discriminator Loss: 1.4481... Generator Loss: 0.6678
Epoch 1/2... Batch 390  Discriminator Loss: 1.5796... Generator Loss: 0.9383
Epoch 1/2... Batch 400  Discriminator Loss: 1.4674... Generator Loss: 0.6494

Epoch 1/2... Batch 410  Discriminator Loss: 1.5500... Generator Loss: 0.5450
Epoch 1/2... Batch 420  Discriminator Loss: 1.4625... Generator Loss: 0.9100
Epoch 1/2... Batch 430  Discriminator Loss: 1.2867... Generator Loss: 0.8892
Epoch 1/2... Batch 440  Discriminator Loss: 1.3729... Generator Loss: 0.6358
Epoch 1/2... Batch 450  Discriminator Loss: 1.2963... Generator Loss: 0.9808
Epoch 1/2... Batch 460  Discriminator Loss: 1.4967... Generator Loss: 0.6870
Epoch 1/2... Batch 470  Discriminator Loss: 1.3654... Generator Loss: 0.9550
Epoch 1/2... Batch 480  Discriminator Loss: 1.4465... Generator Loss: 0.8229
Epoch 1/2... Batch 490  Discriminator Loss: 1.4017... Generator Loss: 1.0538
Epoch 1/2... Batch 500  Discriminator Loss: 1.3583... Generator Loss: 0.9896

Epoch 1/2... Batch 510  Discriminator Loss: 1.4222... Generator Loss: 1.0621
Epoch 1/2... Batch 520  Discriminator Loss: 1.4090... Generator Loss: 0.5954
Epoch 1/2... Batch 530  Discriminator Loss: 1.3977... Generator Loss: 1.1705
Epoch 1/2... Batch 540  Discriminator Loss: 1.3752... Generator Loss: 0.8506
Epoch 1/2... Batch 550  Discriminator Loss: 1.4866... Generator Loss: 1.1832
Epoch 1/2... Batch 560  Discriminator Loss: 1.4226... Generator Loss: 0.7932
Epoch 1/2... Batch 570  Discriminator Loss: 1.5518... Generator Loss: 0.5451
Epoch 1/2... Batch 580  Discriminator Loss: 1.3644... Generator Loss: 1.2490
Epoch 1/2... Batch 590  Discriminator Loss: 1.4470... Generator Loss: 0.6646
Epoch 1/2... Batch 600  Discriminator Loss: 1.2464... Generator Loss: 1.0372

Epoch 1/2... Batch 610  Discriminator Loss: 1.3468... Generator Loss: 0.7565
Epoch 1/2... Batch 620  Discriminator Loss: 1.2948... Generator Loss: 0.8812
Epoch 1/2... Batch 630  Discriminator Loss: 1.3305... Generator Loss: 0.8032
Epoch 1/2... Batch 640  Discriminator Loss: 1.3082... Generator Loss: 1.0318
Epoch 1/2... Batch 650  Discriminator Loss: 1.3843... Generator Loss: 0.8220
Epoch 1/2... Batch 660  Discriminator Loss: 1.5766... Generator Loss: 0.5611
Epoch 1/2... Batch 670  Discriminator Loss: 1.3369... Generator Loss: 0.6561
Epoch 1/2... Batch 680  Discriminator Loss: 1.4229... Generator Loss: 0.7487
Epoch 1/2... Batch 690  Discriminator Loss: 1.3079... Generator Loss: 0.8618
Epoch 1/2... Batch 700  Discriminator Loss: 1.4759... Generator Loss: 1.0133

Epoch 1/2... Batch 710  Discriminator Loss: 1.4139... Generator Loss: 0.6344
Epoch 1/2... Batch 720  Discriminator Loss: 1.4945... Generator Loss: 0.6736
Epoch 1/2... Batch 730  Discriminator Loss: 1.3985... Generator Loss: 1.0230
Epoch 1/2... Batch 740  Discriminator Loss: 1.3291... Generator Loss: 1.2151
Epoch 1/2... Batch 750  Discriminator Loss: 1.3812... Generator Loss: 0.9332
Epoch 1/2... Batch 760  Discriminator Loss: 1.3103... Generator Loss: 0.7235
Epoch 1/2... Batch 770  Discriminator Loss: 1.4938... Generator Loss: 0.6050
Epoch 1/2... Batch 780  Discriminator Loss: 1.3482... Generator Loss: 0.8679
Epoch 1/2... Batch 790  Discriminator Loss: 1.3275... Generator Loss: 0.9760
Epoch 1/2... Batch 800  Discriminator Loss: 1.5106... Generator Loss: 0.5249

Epoch 1/2... Batch 810  Discriminator Loss: 1.4172... Generator Loss: 0.6712
Epoch 1/2... Batch 820  Discriminator Loss: 1.3336... Generator Loss: 0.7340
Epoch 1/2... Batch 830  Discriminator Loss: 1.3987... Generator Loss: 1.2491
Epoch 1/2... Batch 840  Discriminator Loss: 1.1503... Generator Loss: 1.2891
Epoch 1/2... Batch 850  Discriminator Loss: 1.3422... Generator Loss: 0.7985
Epoch 1/2... Batch 860  Discriminator Loss: 1.6132... Generator Loss: 0.5247
Epoch 1/2... Batch 870  Discriminator Loss: 1.6160... Generator Loss: 0.9690
Epoch 1/2... Batch 880  Discriminator Loss: 1.2825... Generator Loss: 0.9425
Epoch 1/2... Batch 890  Discriminator Loss: 1.2930... Generator Loss: 1.0428
Epoch 1/2... Batch 900  Discriminator Loss: 1.2603... Generator Loss: 0.8181

Epoch 1/2... Batch 910  Discriminator Loss: 1.3729... Generator Loss: 0.8209
Epoch 1/2... Batch 920  Discriminator Loss: 1.3569... Generator Loss: 0.9890
Epoch 1/2... Batch 930  Discriminator Loss: 1.4524... Generator Loss: 0.6276
Epoch 2/2... Batch 10  Discriminator Loss: 1.3067... Generator Loss: 0.9549
Epoch 2/2... Batch 20  Discriminator Loss: 1.4887... Generator Loss: 0.5381
Epoch 2/2... Batch 30  Discriminator Loss: 1.3426... Generator Loss: 0.7347
Epoch 2/2... Batch 40  Discriminator Loss: 1.3203... Generator Loss: 0.7249
Epoch 2/2... Batch 50  Discriminator Loss: 1.3889... Generator Loss: 0.9330
Epoch 2/2... Batch 60  Discriminator Loss: 1.4431... Generator Loss: 0.9428
Epoch 2/2... Batch 70  Discriminator Loss: 1.4395... Generator Loss: 0.5695
Epoch 2/2... Batch 80  Discriminator Loss: 1.3897... Generator Loss: 0.8041
Epoch 2/2... Batch 90  Discriminator Loss: 1.3478... Generator Loss: 0.9324
Epoch 2/2... Batch 100  Discriminator Loss: 1.4356... Generator Loss: 0.6739

调参记录¶

初始参数：

batch_size = 64
z_dim = 100
learning_rate = 0.001
beta1 = 0.5

模型架构（filter =5, stride = 1 or 2）：

Discriminator：28x28x1 ---> 14x14x64 ---> 7x7x256 ---> 7x7x512
Generator: 7x7x512 ---> 7x7x256 ---> 14x14x64 ---> 28x28x1

第一次结果：Discriminator Loss: 0.8251... Generator Loss: 0.9614 效果不是很好，基本形状出来了，但是图片灰蒙蒙的

调整beta1：

0.3 ---> Discriminator Loss: 0.8399... Generator Loss: 0.8489 与0.5的结果接近
0.1 ---> Discriminator Loss: 2.1238... Generator Loss: 3.7789 非常糟糕

调整结构：减小辨别器的大小

Discriminator：28x28x1 ---> 14x14x64 ---> 7x7x128 ---> 7x7x256
保持初始参数 ---> Discriminator Loss: 1.6704... Generator Loss: 0.2734
辨别器的loss从一开始就比生成器的loss就来得大，不太合理，感觉互换了竞争地位，还是换回之前的模型结构

调整学习率：

0.005 ---> Discriminator Loss: 1.1054... Generator Loss: 1.7795
0.0005 ---> Discriminator Loss: 0.9000... Generator Loss: 0.6375
0.0001 ---> Discriminator Loss: 0.6731... Generator Loss: 0.9542

z_dim：

128 ---> Discriminator Loss: 0.1254... Generator Loss: 2.7208 糟糕
增加z_dim会更多的增加noise信息

加入dropout感觉提升也不大 Discriminator Loss: 1.1555... Generator Loss: 0.8900

尝试增加迭代次数到3

感觉差别也不大 Discriminator Loss: 0.6617... Generator Loss: 1.0947

根据审阅老师的建议，增加了卷积网络的权重初始化，增加了d_loss的计算平滑，对数据进行了放缩（摆脱了之前图片灰蒙蒙的样子）

目前的loss为：稳定在Discriminator Loss: 1.2-1.4... Generator Loss: 0.7-0.9 Excellent!

CelebA¶

在 CelebA 上运行你的 GANs 模型。在一般的GPU上运行每次迭代大约需要 20 分钟。你可以运行整个迭代，或者当 GANs 开始产生真实人脸图像时停止它。

batch_size = 32
z_dim = 100
learning_rate = 0.0002
beta1 = 0.5


"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
epochs = 3

celeba_dataset = helper.Dataset('celeba', glob(os.path.join(data_dir, 'img_align_celeba/*.jpg')))
with tf.Graph().as_default():
    loss_d, loss_g = train(epochs, batch_size, z_dim, learning_rate, beta1, celeba_dataset.get_batches,
                           celeba_dataset.shape, celeba_dataset.image_mode)

Epoch 1/3... Batch 10  Discriminator Loss: 1.2115... Generator Loss: 1.5248
Epoch 1/3... Batch 20  Discriminator Loss: 1.9405... Generator Loss: 0.6462
Epoch 1/3... Batch 30  Discriminator Loss: 3.6731... Generator Loss: 0.2453
Epoch 1/3... Batch 40  Discriminator Loss: 1.7223... Generator Loss: 0.6057
Epoch 1/3... Batch 50  Discriminator Loss: 2.3389... Generator Loss: 0.1931
Epoch 1/3... Batch 60  Discriminator Loss: 1.3297... Generator Loss: 1.6193
Epoch 1/3... Batch 70  Discriminator Loss: 2.8518... Generator Loss: 0.1680
Epoch 1/3... Batch 80  Discriminator Loss: 2.4531... Generator Loss: 0.2130
Epoch 1/3... Batch 90  Discriminator Loss: 1.2981... Generator Loss: 1.4584
Epoch 1/3... Batch 100  Discriminator Loss: 1.6343... Generator Loss: 0.6648

Epoch 1/3... Batch 110  Discriminator Loss: 2.1845... Generator Loss: 0.8512
Epoch 1/3... Batch 120  Discriminator Loss: 1.9541... Generator Loss: 0.5157
Epoch 1/3... Batch 130  Discriminator Loss: 1.9722... Generator Loss: 0.5076
Epoch 1/3... Batch 140  Discriminator Loss: 0.9135... Generator Loss: 1.8038
Epoch 1/3... Batch 150  Discriminator Loss: 1.1755... Generator Loss: 1.3146
Epoch 1/3... Batch 160  Discriminator Loss: 1.6851... Generator Loss: 0.5875
Epoch 1/3... Batch 170  Discriminator Loss: 2.1325... Generator Loss: 0.2475
Epoch 1/3... Batch 180  Discriminator Loss: 2.0633... Generator Loss: 0.2858
Epoch 1/3... Batch 190  Discriminator Loss: 0.5516... Generator Loss: 2.3470
Epoch 1/3... Batch 200  Discriminator Loss: 1.8524... Generator Loss: 0.4235

Epoch 1/3... Batch 210  Discriminator Loss: 1.5055... Generator Loss: 0.7165
Epoch 1/3... Batch 220  Discriminator Loss: 1.3766... Generator Loss: 0.9411
Epoch 1/3... Batch 230  Discriminator Loss: 1.6069... Generator Loss: 0.7454
Epoch 1/3... Batch 240  Discriminator Loss: 1.3536... Generator Loss: 0.9239
Epoch 1/3... Batch 250  Discriminator Loss: 1.1646... Generator Loss: 1.0169
Epoch 1/3... Batch 260  Discriminator Loss: 1.5339... Generator Loss: 0.8095
Epoch 1/3... Batch 270  Discriminator Loss: 1.4510... Generator Loss: 0.7384
Epoch 1/3... Batch 280  Discriminator Loss: 1.9922... Generator Loss: 0.3889
Epoch 1/3... Batch 290  Discriminator Loss: 1.3988... Generator Loss: 0.6076
Epoch 1/3... Batch 300  Discriminator Loss: 1.4691... Generator Loss: 0.6003

Epoch 1/3... Batch 310  Discriminator Loss: 1.8233... Generator Loss: 0.6377
Epoch 1/3... Batch 320  Discriminator Loss: 1.1222... Generator Loss: 2.0573
Epoch 1/3... Batch 330  Discriminator Loss: 1.5125... Generator Loss: 0.8328
Epoch 1/3... Batch 340  Discriminator Loss: 1.5569... Generator Loss: 0.7651
Epoch 1/3... Batch 350  Discriminator Loss: 1.5277... Generator Loss: 0.7475
Epoch 1/3... Batch 360  Discriminator Loss: 1.7685... Generator Loss: 0.8820
Epoch 1/3... Batch 370  Discriminator Loss: 1.5610... Generator Loss: 0.5346
Epoch 1/3... Batch 380  Discriminator Loss: 1.5669... Generator Loss: 0.8540
Epoch 1/3... Batch 390  Discriminator Loss: 1.8144... Generator Loss: 0.6676
Epoch 1/3... Batch 400  Discriminator Loss: 1.5360... Generator Loss: 0.6488

Epoch 1/3... Batch 410  Discriminator Loss: 1.3761... Generator Loss: 0.8338
Epoch 1/3... Batch 420  Discriminator Loss: 1.3720... Generator Loss: 0.8423
Epoch 1/3... Batch 430  Discriminator Loss: 1.3440... Generator Loss: 0.8140
Epoch 1/3... Batch 440  Discriminator Loss: 1.4375... Generator Loss: 0.9776
Epoch 1/3... Batch 450  Discriminator Loss: 1.5140... Generator Loss: 0.6799
Epoch 1/3... Batch 460  Discriminator Loss: 1.3141... Generator Loss: 0.7173
Epoch 1/3... Batch 470  Discriminator Loss: 1.4957... Generator Loss: 0.6676
Epoch 1/3... Batch 480  Discriminator Loss: 1.1672... Generator Loss: 1.1536
Epoch 1/3... Batch 490  Discriminator Loss: 1.4209... Generator Loss: 1.0291
Epoch 1/3... Batch 500  Discriminator Loss: 1.5152... Generator Loss: 0.6822

Epoch 1/3... Batch 510  Discriminator Loss: 1.3872... Generator Loss: 0.9097
Epoch 1/3... Batch 520  Discriminator Loss: 1.3335... Generator Loss: 0.8283
Epoch 1/3... Batch 530  Discriminator Loss: 1.4014... Generator Loss: 0.9294
Epoch 1/3... Batch 540  Discriminator Loss: 1.5751... Generator Loss: 0.5025
Epoch 1/3... Batch 550  Discriminator Loss: 1.6162... Generator Loss: 0.6402
Epoch 1/3... Batch 560  Discriminator Loss: 1.4570... Generator Loss: 0.8418
Epoch 1/3... Batch 570  Discriminator Loss: 1.8743... Generator Loss: 0.7050
Epoch 1/3... Batch 580  Discriminator Loss: 1.2654... Generator Loss: 0.7746
Epoch 1/3... Batch 590  Discriminator Loss: 1.4023... Generator Loss: 0.8201
Epoch 1/3... Batch 600  Discriminator Loss: 1.4003... Generator Loss: 0.9353

Epoch 1/3... Batch 610  Discriminator Loss: 1.2960... Generator Loss: 0.8946
Epoch 1/3... Batch 620  Discriminator Loss: 1.2364... Generator Loss: 0.7683
Epoch 1/3... Batch 630  Discriminator Loss: 1.1433... Generator Loss: 1.4115
Epoch 1/3... Batch 640  Discriminator Loss: 1.7714... Generator Loss: 0.6413
Epoch 1/3... Batch 650  Discriminator Loss: 1.4730... Generator Loss: 0.8012
Epoch 1/3... Batch 660  Discriminator Loss: 1.2733... Generator Loss: 1.6777
Epoch 1/3... Batch 670  Discriminator Loss: 1.0288... Generator Loss: 1.2907
Epoch 1/3... Batch 680  Discriminator Loss: 1.4785... Generator Loss: 1.4629
Epoch 1/3... Batch 690  Discriminator Loss: 1.5512... Generator Loss: 0.7885
Epoch 1/3... Batch 700  Discriminator Loss: 1.3005... Generator Loss: 1.0756

Epoch 1/3... Batch 710  Discriminator Loss: 1.1415... Generator Loss: 1.1558
Epoch 1/3... Batch 720  Discriminator Loss: 1.5284... Generator Loss: 1.1594
Epoch 1/3... Batch 730  Discriminator Loss: 1.2542... Generator Loss: 1.0307
Epoch 1/3... Batch 740  Discriminator Loss: 1.5477... Generator Loss: 0.6311
Epoch 1/3... Batch 750  Discriminator Loss: 1.5123... Generator Loss: 0.6482
Epoch 1/3... Batch 760  Discriminator Loss: 1.6779... Generator Loss: 0.5824
Epoch 1/3... Batch 770  Discriminator Loss: 1.6002... Generator Loss: 0.7740
Epoch 1/3... Batch 780  Discriminator Loss: 1.3952... Generator Loss: 1.1074
Epoch 1/3... Batch 790  Discriminator Loss: 0.9805... Generator Loss: 1.4949
Epoch 1/3... Batch 800  Discriminator Loss: 1.3628... Generator Loss: 0.8112

Epoch 1/3... Batch 810  Discriminator Loss: 1.4831... Generator Loss: 0.8823
Epoch 1/3... Batch 820  Discriminator Loss: 1.5711... Generator Loss: 0.7207
Epoch 1/3... Batch 830  Discriminator Loss: 1.5298... Generator Loss: 0.6622
Epoch 1/3... Batch 840  Discriminator Loss: 1.3070... Generator Loss: 0.9274
Epoch 1/3... Batch 850  Discriminator Loss: 1.5108... Generator Loss: 1.0278
Epoch 1/3... Batch 860  Discriminator Loss: 1.5795... Generator Loss: 0.7446
Epoch 1/3... Batch 870  Discriminator Loss: 1.5483... Generator Loss: 1.1663
Epoch 1/3... Batch 880  Discriminator Loss: 1.7073... Generator Loss: 0.5900
Epoch 1/3... Batch 890  Discriminator Loss: 1.3564... Generator Loss: 0.7881
Epoch 1/3... Batch 900  Discriminator Loss: 1.2896... Generator Loss: 1.0580

Epoch 1/3... Batch 910  Discriminator Loss: 1.1108... Generator Loss: 1.8433
Epoch 1/3... Batch 920  Discriminator Loss: 1.5374... Generator Loss: 0.6690
Epoch 1/3... Batch 930  Discriminator Loss: 1.2722... Generator Loss: 0.6709
Epoch 1/3... Batch 940  Discriminator Loss: 1.5990... Generator Loss: 0.7263
Epoch 1/3... Batch 950  Discriminator Loss: 1.5372... Generator Loss: 0.8007
Epoch 1/3... Batch 960  Discriminator Loss: 1.3975... Generator Loss: 0.8384
Epoch 1/3... Batch 970  Discriminator Loss: 1.1756... Generator Loss: 0.9768
Epoch 1/3... Batch 980  Discriminator Loss: 1.4310... Generator Loss: 0.8622
Epoch 1/3... Batch 990  Discriminator Loss: 1.5833... Generator Loss: 0.8628
Epoch 1/3... Batch 1000  Discriminator Loss: 1.3363... Generator Loss: 0.9547

from bokeh.plotting import figure, show
from bokeh.io import push_notebook, show, output_notebook
output_notebook()

x_range = [i for i in range(len(loss_d))]
x_range_large = [i for i in range(len(loss_d)) if i % 50 == 0]
loss_d_large = [loss_d[i] for i in x_range_large]
loss_g_large = [loss_g[i] for i in x_range_large]
p = figure(tools = 'save', title='loss_beat1')
p.line(x_range, loss_d, color='#A6CEE3', legend='loss_d')
p.line(x_range, loss_g, color='#FB9A99', legend='loss_g')
p.line(x_range_large, loss_d_large, color='red', legend='loss_d_50')
p.line(x_range_large, loss_g_large, color='blue', legend='loss_g_50')
p.legend.location = "top_left"
show(p)

import numpy as np
print('loss_d mean: ', np.mean(loss_d))
print('loss_g mean: ', np.mean(loss_g))
print('loss_d std: ', np.std(loss_d))
print('loss_g std: ', np.std(loss_g))

loss_d mean:  1.40156
loss_g mean:  0.779231
loss_d std:  0.11954
loss_g std:  0.148314

调参记录¶

更改了模型结构：

Generator: 7x7x512 ---> 14x14x256 ---> 28x28x128 ---> 28x28x3
Discriminator：28x28x3 ---> 14x14x64 ---> 7x7x128 ---> 7x7x256
学习率 0.00025 ---> Discriminator Loss: 0.4439... Generator Loss: 1.4835 感觉很一般，在2000个batch后，g_loss突然上升了

尝试加入dropout：Discriminator Loss: 1.3863... Generator Loss: 0.4561

最后的模型结构：

Generator: 7x7x512 ---> 14x14x256 ---> 28x28x128 ---> 28x28x3
Discriminator：28x28x3 ---> 14x14x64 ---> 7x7x128 ---> 4x4x256

为了查看下模型的最终效果，将迭代次数调整到2次，效果不错，基本看到了人的五官

迭代次数继续增加，应该会得到更好的结果

修改了一些细节，再次训练，为了加快运行，将学习率调整为0.001

batch_size调整：

32 ---> Discriminator Loss: 1.4156... Generator Loss: 0.7807
16 ---> loss没有太多的差异，但是感觉没有32的清晰

beta1调整：

0.4 --->
0.3 ---> 感觉看不出什么差异，决定把loss值画一个折线图

loss平均值

0.5 ---> loss_d:1.47411, loss_g:0.782833
0.4 ---> loss_d:1.46574, loss_g:0.763507
0.3 ---> loss_d:1.47147, loss_g:0.761641 在第3000个batch出现异常值
感觉差别不大，仍旧沿用beta1参数为0.5

最后来一次epoch=3，学习率为0.0002的，可以看到比较清晰的人脸图像了

提交项目¶

提交本项目前，确保运行所有 cells 后保存该文件。

保存该文件为 "dlnd_face_generation.ipynb"，并另存为 HTML 格式 "File" -> "Download as"。提交项目时请附带 "helper.py" 和 "problem_unittests.py" 文件。